Academic Commons

Theses Master's

State Estimation in Lithium-ion Batteries Using Pulse Perturbation and Feedforward Neural Networks

Li, Alan Gen

Predicting battery stored charge, available capacity, and peak power quickly and accurately is important for understanding pack performance and stability. It is proposed that a feedforward neural network (FNN) can estimate this information using cell voltage response to an injected current. Voltage response varies with the internal chemistry, represented by charge, capacity, and impedance. These characteristics are quantified here using state of charge (SoC), state of energy (SoE), and state of power (SoP). Cell response data is collected for various states at constant temperature, resulting in 234 unique voltage responses for training and evaluating the FNN. Training is performed using 3 distinct variations on the data: (1) the full voltage response, (2) individual portions of the response, such as charging or relaxation periods, and (3) fractions of the charge and discharge periods ranging from one-half to a single open-circuit voltage measurement. Using the full response, the average mean absolute error (MAE) is 0.0057 for SoE estimation. The average MAE is below 0.0080 for SoC and SoP estimation. The results for pulse portions show that Charge-rest or Discharge-rest responses perform almost as well as the full pulse. This may inform future pulse design for further optimization. The results for pulse fractions show that error increases as the amount of input data decreases, which validates the hypothesis that pulse perturbation yields high performance in FNN. The technique can be expanded to other temperatures, with potential for estimation of other states, and even degradation mechanisms. Estimation requires 3 minutes of voltage and current data, with no charging history needed and low computational complexity. The proposed method is thus suitable for development of advanced battery management systems in electric vehicles.


  • thumnail for Li 2020 MS Thesis - final.pdf Li 2020 MS Thesis - final.pdf application/pdf 2.73 MB Download File

More About This Work

Academic Units
Electrical Engineering
Thesis Advisors
Preindl, Matthias
M.S., Columbia University
Published Here
August 27, 2020